Merge branch 'main' into dev/asolovev_mkl_support

.ci/env/apt.sh

@@ -32,7 +32,7 @@ function add_repo {
 }
 
 function install_dpcpp {
-    sudo apt-get install -y intel-oneapi-compiler-dpcpp-cpp-2024.1
+    sudo apt-get install -y intel-oneapi-compiler-dpcpp-cpp-2024.2
     sudo bash -c 'echo libintelocl.so > /etc/OpenCL/vendors/intel-cpu.icd'
 }
 
@@ -63,7 +63,7 @@ function install_qemu_emulation_apt {
 }
 
 function install_qemu_emulation_deb {
-    qemu_deb=qemu-user-static_8.2.1+ds-1~bpo12+1_amd64.deb
+    qemu_deb=qemu-user-static_9.0.1+ds-1~bpo12+1_amd64.deb
     set -eo pipefail
     wget http://ftp.debian.org/debian/pool/main/q/qemu/${qemu_deb}
     sudo dpkg -i ${qemu_deb}

.ci/env/environment.yml

@@ -1,7 +1,6 @@
 name: ci-env
 channels:
   - conda-forge
-  - intel
   - defaults
 dependencies:
-  - impi-devel=2021.10.0
+  - impi-devel=2021.12.0

.ci/pipeline/ci.yml

@@ -583,7 +583,7 @@ jobs:
     displayName: 'System info'
   - script: |
       conda update -y -q conda
-      conda create -q -y -n CB -c intel python=$(python.version) tbb mpich
+      conda create -q -y -n CB -c conda-forge python=$(python.version) tbb mpich
     displayName: 'Conda create'
   - script: |
       git clone https://github.com/intel/scikit-learn-intelex.git daal4py

.github/CODEOWNERS

@@ -1,6 +1,6 @@
 # Documentation owners and reviewers
-/docs/         @Vika-F @maria-Petrova @Alexsandruss @aepanchi
-*.md           @Vika-F @maria-Petrova @Alexsandruss @aepanchi
+/docs/         @Vika-F @maria-Petrova @Alexsandruss @bdmoore1
+*.md           @Vika-F @maria-Petrova @Alexsandruss @bdmoore1
 
 # TTP files
 third-party*   @maria-Petrova
@@ -22,7 +22,7 @@ deploy/        @Alexsandruss @napetrov @homksei @ahuber21 @ethanglaser
 dev/           @Alexsandruss @napetrov @homksei @ahuber21 @ethanglaser 
 
 # C++ code
-cpp/           @Alexsandruss @samir-nasibli @KulikovNikita @Alexandr-Solovev
+cpp/           @Alexsandruss @samir-nasibli @Alexandr-Solovev
 
 # Tree based methods
 dtrees         @razdoburdin @ahuber21 @avolkov-intel @icfaust

README.md

@@ -19,7 +19,11 @@
 
 [Installation](#installation)   |   [Documentation](#documentation)   |   [Support](#support)   |   [Examples](#examples)   |   [How to Contribute](CONTRIBUTING.md)   
 
-[![Build Status](https://dev.azure.com/daal/DAAL/_apis/build/status/oneapi-src.oneDAL?branchName=main)](https://dev.azure.com/daal/DAAL/_build/latest?definitionId=5&branchName=main) [![License](https://img.shields.io/github/license/oneapi-src/oneDAL.svg)](https://github.com/oneapi-src/oneDAL/blob/main/LICENSE) [![OpenSSF Best Practices](https://www.bestpractices.dev/projects/8859/badge)](https://www.bestpractices.dev/projects/8859) [![Join the community on GitHub Discussions](https://badgen.net/badge/join%20the%20discussion/on%20github/black?icon=github)](https://github.com/oneapi-src/oneDAL/discussions)
+[![Build Status](https://dev.azure.com/daal/DAAL/_apis/build/status/oneapi-src.oneDAL?branchName=main)](https://dev.azure.com/daal/DAAL/_build/latest?definitionId=5&branchName=main)
+[![License](https://img.shields.io/github/license/oneapi-src/oneDAL.svg)](https://github.com/oneapi-src/oneDAL/blob/main/LICENSE)
+[![OpenSSF Best Practices](https://www.bestpractices.dev/projects/8859/badge)](https://www.bestpractices.dev/projects/8859)
+[![OpenSSF Scorecard](https://api.securityscorecards.dev/projects/github.com/oneapi-src/oneDAL/badge)](https://securityscorecards.dev/viewer/?uri=github.com/oneapi-src/oneDAL)
+[![Join the community on GitHub Discussions](https://badgen.net/badge/join%20the%20discussion/on%20github/black?icon=github)](https://github.com/oneapi-src/oneDAL/discussions)
 
 oneAPI Data Analytics Library (oneDAL) is a powerful machine learning library that helps you accelerate big data analysis at all stages: **preprocessing**, **transformation**, **analysis**, **modeling**, **validation**, and **decision making**.
 

SECURITY.md

@@ -19,13 +19,13 @@
 
 As an open-source project, we understand the importance of and responsibility
 for security. This Security Policy outlines our guidelines and procedures to
-ensure the highest level of security and trust for Intel(R) oneAPI Data Analytics Library.
+ensure the highest level of security and trust for oneDAL
 
 ## Supported Versions
 
 Security vulnerabilities are fixed in the [latest version][1]
 and delivered as a patch release. We don't guarantee security fixes to be
-back-ported to older Intel(R) oneAPI Data Analytics Library versions.
+back-ported to older oneDAL versions.
 
 ## Report a Vulnerability
 
@@ -52,24 +52,24 @@ Along with the report, provide the following info:
 
 ### When Should I Report a Vulnerability?
 
-* You think you discovered a potential security vulnerability in Intel(R) oneAPI Data Analytics Library.
-* You are unsure how the potential vulnerability affects Intel(R) oneAPI Data Analytics Library.
+* You think you discovered a potential security vulnerability in oneDAL
+* You are unsure how the potential vulnerability affects oneDAL
 * You think you discovered a vulnerability in another project or 3rd party
-component on which Intel(R) oneAPI Data Analytics Library depends. If the issue is not fixed in the 3rd party
+component on which oneDAL depends. If the issue is not fixed in the 3rd party
 component, try to report directly there first.
 
 ### When Should I NOT Report a Vulnerability?
 
 * You got an automated scan hit and are unable to provide details.
-* You need help using Intel(R) oneAPI Data Analytics Library for security.
+* You need help using oneDAL for security.
 * You need help applying security-related updates.
 * Your issue is not security-related.
 
 ## Security Reports Review Process
 
 We aim to respond quickly to your inquiry and coordinate a fix and
 disclosure with you. All confirmed security vulnerabilities will be addressed
-according to severity level and impact on Intel(R) oneAPI Data Analytics Library. Normally, security issues are fixed in the next planned release.
+according to severity level and impact on oneDAL Normally, security issues are fixed in the next planned release.
 
 ## Disclosure Policy
 
@@ -88,4 +88,6 @@ If you have any suggestions on how this Policy could be improved, submit
 an issue or a pull request to this repository. **Do not** report
 potential vulnerabilities or security flaws via a pull request.
 
-[1]: https://github.com/oneapi-src/oneDAL/releases
+[1]: https://github.com/oneapi-src/oneDAL/releases
+[2]: https://github.com/oneapi-src/oneDAL/security/advisories/new
+[3]: https://github.com/oneapi-src/oneDAL/security/advisories

WORKSPACE

@@ -100,8 +100,8 @@ http_archive(
 
 http_archive(
     name = "fmt",
-    url = "https://github.com/fmtlib/fmt/archive/10.2.1.tar.gz",
-    sha256 = "1250e4cc58bf06ee631567523f48848dc4596133e163f02615c97f78bab6c811",
-    strip_prefix = "fmt-10.2.1",
+    url = "https://github.com/fmtlib/fmt/archive/11.0.2.tar.gz",
+    sha256 = "6cb1e6d37bdcb756dbbe59be438790db409cdb4868c66e888d5df9f13f7c027f",
+    strip_prefix = "fmt-11.0.2",
     build_file = "@onedal//dev/bazel/deps:fmt.tpl.BUILD",
 )

cpp/daal/include/services/env_detect.h

@@ -198,6 +198,10 @@ class DAAL_EXPORT Environment : public Base
     void initNumberOfThreads();
 
     env _env;
+    // Pointer to the oneapi::tbb::task_scheduler_handle class object, global for oneDAL.
+    // The oneapi::tbb::task_scheduler_handle and the oneapi::tbb::finalize function
+    // allow user to wait for completion of worker threads.
+    void * _schedulerHandle;
     void * _globalControl;
     SharedPtr<services::internal::sycl::ExecutionContextIface> _executionContext;
 };

cpp/daal/src/externals/core_threading_win_dll.cpp

@@ -143,7 +143,9 @@ typedef void (*_daal_wait_task_group_t)(void * taskGroupPtr);
 
 typedef bool (*_daal_is_in_parallel_t)();
 typedef void (*_daal_tbb_task_scheduler_free_t)(void *& globalControl);
+typedef void (*_daal_tbb_task_scheduler_handle_free_t)(void *& schedulerHandle);
 typedef size_t (*_setNumberOfThreads_t)(const size_t, void **);
+typedef size_t (*_setSchedulerHandle_t)(void **);
 typedef void * (*_daal_threader_env_t)();
 
 typedef void (*_daal_parallel_sort_int32_t)(int *, int *);
@@ -205,10 +207,12 @@ static _daal_del_task_group_t _daal_del_task_group_ptr   = NULL;
 static _daal_run_task_group_t _daal_run_task_group_ptr   = NULL;
 static _daal_wait_task_group_t _daal_wait_task_group_ptr = NULL;
 
-static _daal_is_in_parallel_t _daal_is_in_parallel_ptr                   = NULL;
-static _daal_tbb_task_scheduler_free_t _daal_tbb_task_scheduler_free_ptr = NULL;
-static _setNumberOfThreads_t _setNumberOfThreads_ptr                     = NULL;
-static _daal_threader_env_t _daal_threader_env_ptr                       = NULL;
+static _daal_is_in_parallel_t _daal_is_in_parallel_ptr                                 = NULL;
+static _daal_tbb_task_scheduler_free_t _daal_tbb_task_scheduler_free_ptr               = NULL;
+static _daal_tbb_task_scheduler_handle_free_t _daal_tbb_task_scheduler_handle_free_ptr = NULL;
+static _setNumberOfThreads_t _setNumberOfThreads_ptr                                   = NULL;
+static _setSchedulerHandle_t _setSchedulerHandle_ptr                                   = NULL;
+static _daal_threader_env_t _daal_threader_env_ptr                                     = NULL;
 
 static _daal_parallel_sort_int32_t _daal_parallel_sort_int32_ptr                         = NULL;
 static _daal_parallel_sort_uint64_t _daal_parallel_sort_uint64_ptr                       = NULL;
@@ -657,6 +661,16 @@ DAAL_EXPORT void _daal_tbb_task_scheduler_free(void *& init)
     return _daal_tbb_task_scheduler_free_ptr(init);
 }
 
+DAAL_EXPORT void _daal_tbb_task_scheduler_handle_free(void *& init)
+{
+    load_daal_thr_dll();
+    if (_daal_tbb_task_scheduler_handle_free_ptr == NULL)
+    {
+        _daal_tbb_task_scheduler_handle_free_ptr = (_daal_tbb_task_scheduler_handle_free_t)load_daal_thr_func("_daal_tbb_task_scheduler_handle_free");
+    }
+    return _daal_tbb_task_scheduler_handle_free_ptr(init);
+}
+
 DAAL_EXPORT size_t _setNumberOfThreads(const size_t numThreads, void ** init)
 {
     load_daal_thr_dll();
@@ -667,6 +681,16 @@ DAAL_EXPORT size_t _setNumberOfThreads(const size_t numThreads, void ** init)
     return _setNumberOfThreads_ptr(numThreads, init);
 }
 
+DAAL_EXPORT size_t _setSchedulerHandle(void ** init)
+{
+    load_daal_thr_dll();
+    if (_setSchedulerHandle_ptr == NULL)
+    {
+        _setSchedulerHandle_ptr = (_setSchedulerHandle_t)load_daal_thr_func("_setSchedulerHandle");
+    }
+    return _setSchedulerHandle_ptr(init);
+}
+
 DAAL_EXPORT void * _daal_threader_env()
 {
     load_daal_thr_dll();

cpp/daal/src/externals/service_stat_ref.h

@@ -1,6 +1,7 @@
 /* file: service_stat_ref.h */
 /*******************************************************************************
 * Copyright 2023 Intel Corporation
+* Copyright contributors to the oneDAL project
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
@@ -24,8 +25,9 @@
 #ifndef __SERVICE_STAT_REF_H__
 #define __SERVICE_STAT_REF_H__
 
-typedef void (*func_type)(DAAL_INT, DAAL_INT, DAAL_INT, void *);
+#include "src/externals/service_memory.h"
 
+typedef void (*func_type)(DAAL_INT, DAAL_INT, DAAL_INT, void *);
 extern "C"
 {
 #define __DAAL_VSL_SS_MATRIX_STORAGE_COLS           0x00020000
@@ -194,8 +196,37 @@ struct RefStatistics<double, cpu>
 
     static int x2c_mom(const double * data, const __int64 nFeatures, const __int64 nVectors, double * variance, const __int64 method)
     {
-        int errcode = 0;
-
+        // E(x-\mu)^2 = E(x^2) - \mu^2
+        int errcode  = 0;
+        double * sum = (double *)daal::services::internal::service_calloc<double, cpu>(nFeatures, sizeof(double));
+        if (!sum) return -4;
+        daal::services::internal::service_memset<double, cpu>(variance, double(0), nFeatures);
+        DAAL_INT feature_ptr, vec_ptr;
+        double wtInv      = (double)1 / nVectors;
+        double wtInvMinus = (double)1 / (nVectors - 1);
+        double pt         = 0;
+        for (vec_ptr = 0; vec_ptr < nVectors; ++vec_ptr)
+        {
+#pragma omp simd
+            for (feature_ptr = 0; feature_ptr < nFeatures; ++feature_ptr)
+            {
+                pt = data[vec_ptr * nFeatures + feature_ptr];
+                sum[feature_ptr] += pt;
+                variance[feature_ptr] += (pt * pt); // 2RSum
+            }
+        }
+        double sumSqDivN; // S^2/n = n*\mu^2
+#pragma omp simd
+        for (feature_ptr = 0; feature_ptr < nFeatures; ++feature_ptr)
+        {
+            sumSqDivN = sum[feature_ptr];
+            sumSqDivN *= sumSqDivN;
+            sumSqDivN *= wtInv;
+            variance[feature_ptr] -= sumSqDivN; // (2RSum-S^2/n)
+            variance[feature_ptr] *= wtInvMinus;
+        }
+        daal::services::internal::service_free<double, cpu>(sum);
+        sum = NULL;
         return errcode;
     }
 
@@ -276,8 +307,37 @@ struct RefStatistics<float, cpu>
 
     static int x2c_mom(const float * data, const __int64 nFeatures, const __int64 nVectors, float * variance, const __int64 method)
     {
+        // E(x-\mu)^2 = E(x^2) - \mu^2
         int errcode = 0;
-
+        float * sum = (float *)daal::services::internal::service_calloc<float, cpu>(nFeatures, sizeof(float));
+        if (!sum) return -4;
+        daal::services::internal::service_memset<float, cpu>(variance, float(0), nFeatures);
+        DAAL_INT feature_ptr, vec_ptr;
+        float wtInv      = (float)1 / nVectors;
+        float wtInvMinus = (float)1 / (nVectors - 1);
+        float pt         = 0;
+        for (vec_ptr = 0; vec_ptr < nVectors; ++vec_ptr)
+        {
+#pragma omp simd
+            for (feature_ptr = 0; feature_ptr < nFeatures; ++feature_ptr)
+            {
+                pt = data[vec_ptr * nFeatures + feature_ptr];
+                sum[feature_ptr] += pt;
+                variance[feature_ptr] += (pt * pt); // 2RSum
+            }
+        }
+        float sumSqDivN; // S^2/n = n*\mu^2
+#pragma omp simd
+        for (feature_ptr = 0; feature_ptr < nFeatures; ++feature_ptr)
+        {
+            sumSqDivN = sum[feature_ptr];
+            sumSqDivN *= sumSqDivN;
+            sumSqDivN *= wtInv;
+            variance[feature_ptr] -= sumSqDivN; // (2RSum-S^2/n)
+            variance[feature_ptr] *= wtInvMinus;
+        }
+        daal::services::internal::service_free<float, cpu>(sum);
+        sum = NULL;
         return errcode;
     }
 

cpp/daal/src/services/env_detect.cpp

@@ -125,7 +125,7 @@ DAAL_EXPORT void daal::services::Environment::setDynamicLibraryThreadingTypeOnWi
     initNumberOfThreads();
 }
 
-DAAL_EXPORT daal::services::Environment::Environment() : _globalControl {}
+DAAL_EXPORT daal::services::Environment::Environment() : _schedulerHandle {}, _globalControl {}
 {
     _env.cpuid_init_flag = false;
     _env.cpuid           = -1;
@@ -137,7 +137,14 @@ DAAL_EXPORT daal::services::Environment::Environment(const Environment & e) : da
 DAAL_EXPORT void daal::services::Environment::initNumberOfThreads()
 {
     if (isInit) return;
-
+        // Initializes global oneapi::tbb::task_scheduler_handle object in oneDAL to prevent the unexpected
+        // destruction of the calling thread.
+        // When the oneapi::tbb::finalize function is called with an oneapi::tbb::task_scheduler_handle
+        // instance, it blocks the calling thread until the completion of all worker
+        // threads that were implicitly created by the library.
+#if defined(TARGET_X86_64)
+    daal::setSchedulerHandle(&_schedulerHandle);
+#endif
     /* if HT enabled - set _numThreads to physical cores num */
     if (daal::internal::ServiceInst::serv_get_ht())
     {
@@ -156,7 +163,6 @@ DAAL_EXPORT void daal::services::Environment::initNumberOfThreads()
 DAAL_EXPORT daal::services::Environment::~Environment()
 {
     daal::services::daal_free_buffers();
-    _daal_tbb_task_scheduler_free(_globalControl);
 }
 
 void daal::services::Environment::_cpu_detect(int enable)
@@ -171,6 +177,9 @@ void daal::services::Environment::_cpu_detect(int enable)
 DAAL_EXPORT void daal::services::Environment::setNumberOfThreads(const size_t numThreads)
 {
     isInit = true;
+#if defined(TARGET_X86_64)
+    daal::setSchedulerHandle(&_schedulerHandle);
+#endif
     daal::setNumberOfThreads(numThreads, &_globalControl);
 }
 

cpp/daal/src/threading/service_thread_pinner.cpp

@@ -234,6 +234,9 @@ class thread_pinner_impl_t : public tbb::task_scheduler_observer
 thread_pinner_impl_t::thread_pinner_impl_t(void (*read_topo)(int &, int &, int &, int **), void (*deleter)(void *))
     : pinner_arena(nthreads = daal::threader_get_threads_number()), tbb::task_scheduler_observer(pinner_arena), topo_deleter(deleter)
 {
+    #if defined(TARGET_X86_64)
+    pinner_arena.initialize();
+    #endif
     do_pinning = (nthreads > 0) ? true : false;
     is_pinning.set(0);
 

cpp/daal/src/threading/threading.cpp

@@ -61,6 +61,33 @@ DAAL_EXPORT void _daal_tbb_task_scheduler_free(void *& globalControl)
     }
 }
 
+DAAL_EXPORT void _daal_tbb_task_scheduler_handle_free(void *& schedulerHandle)
+{
+    // Note: TBB 13 deletes task_scheduler_handle itself during the destruction of thread context
+
+    // #if defined(TARGET_X86_64)
+    //     if (schedulerHandle)
+    //     {
+    //         delete reinterpret_cast<tbb::task_scheduler_handle *>(schedulerHandle);
+    //         schedulerHandle = nullptr;
+    //     }
+    // #endif
+}
+
+DAAL_EXPORT size_t _setSchedulerHandle(void ** schedulerHandle)
+{
+#if defined(TARGET_X86_64)
+    #if (TBB_INTERFACE_VERSION < 12120)
+    schedulerHandle = nullptr;
+    #else
+    *schedulerHandle = reinterpret_cast<void *>(new tbb::task_scheduler_handle(tbb::attach {}));
+    #endif
+    // It is necessary for initializing tbb in cases where DAAL does not use it.
+    tbb::task_arena {}.initialize();
+#endif
+    return 0;
+}
+
 DAAL_EXPORT size_t _setNumberOfThreads(const size_t numThreads, void ** globalControl)
 {
     static tbb::spin_mutex mt;